4 ms·
Isn't this just saying what we already knew, that tokenization has these fundamental flaws? Did we need a paper for it?
by mvkel 2y ago
Isn't this just saying what we already knew, that tokenization has these fundamental flaws? Did we need a paper for it?
- gs17 2y agoIt is, I'd even say it's just "LLMs aren't very good at math" rather than a "Complete Reasoning Breakdown". Still, evaluating when/how they fail could be useful.